Can artificial intelligence models predict hospital stays following non-small cell lung carcinoma surgery?

Abstract

Introduction: The hospitalization day after non-small cell lung carcinoma surgery is important in terms of medical, financial, and social aspects. Aim: Our study aimed to accurately and successfully predict the rate of postoperative hospitalization shorter than 9 days versus longer than 9 days using artificial intelligence algorithms. Material and methods: The study included 953 patients who underwent surgery for non-small-cell lung carcinoma between 2001 and 2023. The patients' input data consisted of clinical data, laboratory data, respiratory parameters, and radiological and pathological data. The output data included the hospitalization date. We used a fully connected neural network and the k-layer validation method. Results: The algorithm's training data had a sensitivity value of 90.3%, a positive predictive value of 87.8%, and an accuracy value of 82.6%. The algorithm's F1 1 score was 89.0%, the F1 0 value was 58.3%, and the F1 mean score was 73.6% for the training data. For the test data, the algorithm's sensitivity value was 83.8%, the positive predictive value was 88.6%, and the accuracy value was 78.7%. The F1 1 score was 86.1%, the test F1 0 score was 54.5%, and the F1 average score was 70.3%. The algorithm for the test data created a ROC curve with an area under the curve of 0.82 (AUC = 0.82). Conclusions: Artificial intelligence algorithms determined the length of hospital stay after non-small cell lung carcinoma surgery with high accuracy and confidence. During the preoperative period, the estimation of hospital stay length will contribute to personalized patient care.